Patch to handle squeezed output of custom action recognition model

Continuing the discussion from Using Custom action recognition Model in Deepstream 3D action recognition:

Please provide complete information as applicable to your setup.

• Hardware Platform GPU
• DeepStream Version 6.1
• JetPack Version (valid for Jetson only)
• TensorRT Version 8.4
• NVIDIA GPU Driver Version 11.7
• Issue Type new requirements

Hi @AastaLLL

Can you please provide a patch file for the referenced link which can be used for dgpu with Cuda 11.7 for Tesla-P100.

Thanks,
Hemang

I am checking.

you can use the same patch.

Hi @fanzh ,

I get the below error while applying the patch,

error: patch failed: nvdsinfer_func_utils.cpp:98
error: nvdsinfer_func_utils.cpp: patch does not apply

Can you please help me with the same?

Do i have to make any changes in dims.numElements value from nvdsinfer_func_utils.cpp file?

Thanks,
Hemang

you can merge the code manually, please open the patch with txt, copy the code with “+”, to SplitFullDims() of nvdsinfer_func_utils.cpp, then build and copy like this Using Custom action recognition Model in Deepstream 3D action recognition - #12 by AastaLLL

I have checked, It seems the patch was added and still gave the above error. Not sure why.

I have built it and it is working fine now.

Thanks.

Glad to know you fixed it, thanks for the update!